Add constrained span(initializer_list) constructor by rparolin · Pull Request #8314 · NVIDIA/cccl

rparolin · 2026-04-07T17:40:08Z

Summary

Re-adds the span(initializer_list) constructor that P4144R1 removed from C++26, using Hana Dusíková's constrained fix
Constructor is constrained with same_as<value_type> (prevents narrowing) and is_const_v<element_type> (prevents dangling)
Added to both static and dynamic extent specializations
No CTAD deduction guide — it causes regressions in existing braced-init-list deduction (e.g., span s{ptr, ptr})

Motivation

This implementation supports a WG21 paper proposing to restore the span(initializer_list) constructor with safety constraints. Having a working implementation in libcu++ strengthens the paper with real-world evidence.

Key design decisions

Uses initializer_list<_InitListValueType> instead of Hana's original initializer_list<const _InitListValueType> — the const is redundant because initializer_list<T>::begin() already returns const T*, and omitting it simplifies SFINAE fallback deduction
No deduction guide: span(initializer_list<T>) -> span<const T> causes C++ to prefer the initializer_list interpretation for all braced-init-list CTAD, breaking existing span s{ptr, sentinel} deduction

Test plan

initializer_list.pass.cpp — dynamic/static extent, named variables, empty lists, value integrity, bool literals, const volatile
initializer_list.fail.cpp — non-const element type, narrowing (int→bool, int→short, double→int)
Full span test suite: 38/38 pass

🤖 Generated with Claude Code

Re-add the span(initializer_list) constructor that P4144R1 removed from C++26. Uses Hana Dusikova's fix: constrain with same_as<value_type> to prevent narrowing and is_const_v<element_type> to prevent dangling. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…guide Move #include <cuda/std/initializer_list> to the top-level header group (alongside array, cstddef, version) per project include ordering convention. Add CTAD deduction guide: span(initializer_list<T>) -> span<const T>. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Tests cover dynamic/static extent construction, explicit initializer_list variables, empty lists, value integrity, bool literals, const volatile element types, and CTAD deduction. Both constexpr and runtime paths verified. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Test dynamic/static extent construction, explicit initializer_list variable, empty lists, value integrity, bool literals, const volatile element type, and CTAD deduction guide. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Verify that the constrained span(initializer_list) constructor correctly rejects non-const element types and type-mismatched initializer lists. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The span constructed from a temporary braced-init-list dangles after the initializer_list temporary is destroyed. Use named variables so the backing storage outlives element access through the span. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- const volatile: only check size (volatile reads are not constexpr, and volatile int has no operator== with int) - CTAD: use parentheses initialization to avoid brace-init ambiguity Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The libcudacxx lit config only supports .pass.cpp, .fail.cpp, .runfail.cpp, and .sh.cpp suffixes. Rename to .fail.cpp format. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The deduction guide span(initializer_list<T>) -> span<const T> causes C++ to prefer the initializer_list interpretation for braced-init-list CTAD, breaking existing span s{ptr, ptr} deduction (deduces span<const int*> instead of span<int>). Remove the guide; the constrained constructor works fine with explicit types. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

copy-pr-bot · 2026-04-07T17:40:12Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

- Remove redundant type instantiations (long, double) that test the same code path as int - Remove redundant narrowing case (const short, same as const bool) - Add span<int* const> positive test (is_const_v is true for const ptr) - Add span<const int*> negative test (is_const_v is false — the pointer itself is not const, only the pointee is) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

…n.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

…n.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

…lity MSVC 14.29 eagerly evaluates is_const_v<element_type> during class instantiation because element_type is a class template parameter. This causes a hard error instead of SFINAE discard. Adding a defaulted function template parameter _Ep = element_type makes the condition depend on the function's own template parameter, deferring evaluation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

miscco · 2026-04-08T06:55:46Z

Thanks a lot @rparolin for addin the fix

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

github-actions · 2026-04-08T17:39:19Z

🥳 CI Workflow Results

🟩 Finished in 2h 04m: Pass: 100%/108 | Total: 2d 23h | Max: 2h 03m | Hits: 79%/281680

See results here.

* feat: add constrained span(initializer_list) constructor Re-add the span(initializer_list) constructor that P4144R1 removed from C++26. Uses Hana Dusikova's fix: constrain with same_as<value_type> to prevent narrowing and is_const_v<element_type> to prevent dangling. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: move initializer_list include to top-level group, add deduction guide Move #include <cuda/std/initializer_list> to the top-level header group (alongside array, cstddef, version) per project include ordering convention. Add CTAD deduction guide: span(initializer_list<T>) -> span<const T>. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add positive test cases for span(initializer_list) constructor Tests cover dynamic/static extent construction, explicit initializer_list variables, empty lists, value integrity, bool literals, const volatile element types, and CTAD deduction. Both constexpr and runtime paths verified. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add passing tests for span(initializer_list) constructor Test dynamic/static extent construction, explicit initializer_list variable, empty lists, value integrity, bool literals, const volatile element type, and CTAD deduction guide. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: add compile-failure tests for span(initializer_list) constructor Verify that the constrained span(initializer_list) constructor correctly rejects non-const element types and type-mismatched initializer lists. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * style: apply clang-format to initializer_list.verify.cpp Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: use named initializer_list variables to avoid dangling in tests The span constructed from a temporary braced-init-list dangles after the initializer_list temporary is destroyed. Use named variables so the backing storage outlives element access through the span. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: fix const volatile and CTAD tests - const volatile: only check size (volatile reads are not constexpr, and volatile int has no operator== with int) - CTAD: use parentheses initialization to avoid brace-init ambiguity Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: replace .verify.cpp with .fail.cpp for negative tests The libcudacxx lit config only supports .pass.cpp, .fail.cpp, .runfail.cpp, and .sh.cpp suffixes. Rename to .fail.cpp format. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: remove initializer_list deduction guide to avoid CTAD regression The deduction guide span(initializer_list<T>) -> span<const T> causes C++ to prefer the initializer_list interpretation for braced-init-list CTAD, breaking existing span s{ptr, ptr} deduction (deduces span<const int*> instead of span<int>). Remove the guide; the constrained constructor works fine with explicit types. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: remove extra blank line, add dangling comments to size-only tests Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * fix: improve deviation comment to explain why const is redundant Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: audit and refine initializer_list tests - Remove redundant type instantiations (long, double) that test the same code path as int - Remove redundant narrowing case (const short, same as const bool) - Add span<int* const> positive test (is_const_v is true for const ptr) - Add span<const int*> negative test (is_const_v is false — the pointer itself is not const, only the pointee is) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: replace magic numbers with il.size() in size assertions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * test: fix remaining magic numbers in initializer_list variable tests Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * style: apply clang-format alignment Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Update libcudacxx/include/cuda/std/span Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com> * Update libcudacxx/test/libcudacxx/std/containers/views/views.span/span.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com> * Update libcudacxx/include/cuda/std/span Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com> * Update libcudacxx/test/libcudacxx/std/containers/views/views.span/span.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com> * fix: add defaulted _Ep template param for MSVC 14.29 SFINAE compatibility MSVC 14.29 eagerly evaluates is_const_v<element_type> during class instantiation because element_type is a class template parameter. This causes a hard error instead of SFINAE discard. Adding a defaulted function template parameter _Ep = element_type makes the condition depend on the function's own template parameter, deferring evaluation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * style: apply clang-format Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * Update libcudacxx/include/cuda/std/span Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com> --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com> Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

rparolin and others added 12 commits April 7, 2026 09:16

test: add compile-failure tests for span(initializer_list) constructor

5b30088

Verify that the constrained span(initializer_list) constructor correctly rejects non-const element types and type-mismatched initializer lists. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

style: apply clang-format to initializer_list.verify.cpp

6a782a0

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

test: replace .verify.cpp with .fail.cpp for negative tests

5029085

The libcudacxx lit config only supports .pass.cpp, .fail.cpp, .runfail.cpp, and .sh.cpp suffixes. Rename to .fail.cpp format. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

fix: remove extra blank line, add dangling comments to size-only tests

5a061b3

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

fix: improve deviation comment to explain why const is redundant

f86b58a

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

github-project-automation Bot added this to CCCL Apr 7, 2026

github-project-automation Bot moved this to Todo in CCCL Apr 7, 2026

cccl-authenticator-app Bot moved this from Todo to In Progress in CCCL Apr 7, 2026

rparolin and others added 4 commits April 7, 2026 10:49

test: replace magic numbers with il.size() in size assertions

08579fb

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

test: fix remaining magic numbers in initializer_list variable tests

5595d23

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

style: apply clang-format alignment

0122237

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

mhoemmen reviewed Apr 7, 2026

View reviewed changes

Comment thread libcudacxx/test/libcudacxx/std/containers/views/views.span/span.cons/initializer_list.pass.cpp Outdated

mhoemmen reviewed Apr 7, 2026

View reviewed changes

Comment thread libcudacxx/test/libcudacxx/std/containers/views/views.span/span.cons/initializer_list.pass.cpp Outdated

mhoemmen reviewed Apr 7, 2026

View reviewed changes

Comment thread libcudacxx/include/cuda/std/span Outdated

mhoemmen reviewed Apr 7, 2026

View reviewed changes

Comment thread libcudacxx/include/cuda/std/span Outdated

rparolin and others added 4 commits April 7, 2026 13:16

Update libcudacxx/include/cuda/std/span

a14b07d

Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

Update libcudacxx/test/libcudacxx/std/containers/views/views.span/spa…

7f8da90

…n.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

Update libcudacxx/include/cuda/std/span

fb3702a

Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

Update libcudacxx/test/libcudacxx/std/containers/views/views.span/spa…

b9df966

…n.cons/initializer_list.pass.cpp Co-authored-by: Mark Hoemmen <mhoemmen@users.noreply.github.com>

rparolin marked this pull request as ready for review April 7, 2026 20:18

rparolin requested a review from a team as a code owner April 7, 2026 20:18

rparolin requested a review from ericniebler April 7, 2026 20:18

cccl-authenticator-app Bot moved this from In Progress to In Review in CCCL Apr 7, 2026

This comment has been minimized.

Sign in to view

rparolin and others added 2 commits April 7, 2026 15:10

style: apply clang-format

9600171

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

This comment has been minimized.

Sign in to view

miscco approved these changes Apr 8, 2026

View reviewed changes

Comment thread libcudacxx/include/cuda/std/span Outdated

Update libcudacxx/include/cuda/std/span

bcc72db

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

miscco merged commit f10648a into NVIDIA:main Apr 8, 2026
125 of 127 checks passed

github-project-automation Bot moved this from In Review to Done in CCCL Apr 8, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add constrained span(initializer_list) constructor#8314

Add constrained span(initializer_list) constructor#8314
miscco merged 23 commits intoNVIDIA:mainfrom
rparolin:feature/span-initializer-list-ctor

rparolin commented Apr 7, 2026

Uh oh!

copy-pr-bot Bot commented Apr 7, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

Uh oh!

miscco commented Apr 8, 2026

Uh oh!

github-actions Bot commented Apr 8, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

rparolin commented Apr 7, 2026

Summary

Motivation

Key design decisions

Test plan

Uh oh!

copy-pr-bot Bot commented Apr 7, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

Uh oh!

miscco commented Apr 8, 2026

Uh oh!

github-actions Bot commented Apr 8, 2026

🥳 CI Workflow Results

🟩 Finished in 2h 04m: Pass: 100%/108 | Total: 2d 23h | Max: 2h 03m | Hits: 79%/281680

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants